diabolic6045/Sanskrit-Qwen2.5-VL-7B-Instruct-OCR

VISIONPricing:Input $0.4 / Output $0.8Concurrent Unit Cost:1Model Size:7BQuant:FP8Context Size:32kPublished:Sep 8, 2025License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

diabolic6045/Sanskrit-Qwen2.5-VL-7B-Instruct-OCR is a 7 billion parameter vision-language model fine-tuned from Qwen/Qwen2.5-VL-7B-Instruct. It is specifically adapted for Sanskrit Optical Character Recognition (OCR) tasks, converting Sanskrit text from images into machine-readable text. This model leverages LoRA fine-tuning on a specialized Sanskrit OCR parallel corpus, maintaining multimodal understanding while excelling in transcribing various Sanskrit scripts and fonts.

Loading preview...

Sanskrit-Qwen2.5-VL-7B-Instruct-OCR Overview

This model is a specialized 7 billion parameter vision-language model, fine-tuned from the Qwen/Qwen2.5-VL-7B-Instruct base model. Its primary function is to perform Optical Character Recognition (OCR) for Sanskrit text, converting images containing Sanskrit script into digital text. The fine-tuning process utilized LoRA (Low-Rank Adaptation) on a dedicated Sanskrit OCR parallel corpus, enhancing its ability to accurately recognize and transcribe Sanskrit characters and words from visual inputs.

Key Capabilities

  • Sanskrit OCR: Transcribes Sanskrit text directly from images.
  • Multimodal Understanding: Processes both visual (image) and linguistic (text) inputs simultaneously.
  • Script and Font Handling: Capable of recognizing various Sanskrit scripts and fonts.
  • High Character Accuracy: Achieves 86.38% character-level accuracy on its evaluation set, indicating strong individual character recognition.

Good For

  • Digitization of Sanskrit Texts: Ideal for converting historical or printed Sanskrit documents into searchable, editable digital formats.
  • Academic Research: Assisting scholars and researchers working with Sanskrit literature by automating text extraction from images.
  • Language Technology Development: Serving as a foundational component for applications requiring Sanskrit text input from visual sources.
  • Specialized OCR Tasks: When general-purpose OCR models struggle with the complexities of Sanskrit script, this model offers a targeted solution.

Finetunes
1 models